Suede AI Agent StudioDocsStart buildingOpen workspace
Navigation menu

Agents & workflows4 min readUpdated

Add human approval and escalation to your AI agents

Quick answer

Agent Studio puts a person in the loop wherever you want one. Gate any step with a Branch on an approval flag, hold social posts and review replies for your yes, and escalate to your team with Create Task, a note or a Slack alert. Everything routine runs on its own; the judgment calls come to you.

What you'll have

  • An approval gate in front of any step that speaks for your business
  • Automatic hand-offs to the right person when an agent hits a judgment call
  • A clear map of which work runs alone and which waits for you
On this page
  1. Where can I add a human approval step?
  2. How do I build an approval gate into an agent?
  3. How does an AI agent escalate to a human?
  4. What should make an agent escalate?
  5. Questions

Where can I add a human approval step?#

Anywhere an agent speaks for you or spends money. Agent Studio ships approval in four places, and you choose how much runs on its own.

Built-in approval points
WhereHow the approval worksSet it in
Any agent stepA Branch reads an approval field; the true path runs the step, the false path returns for reviewThe builder
Social postsTurn on Needs approval and the post waits in CRM Social for your yesThe social publish step
Critical reviewsAI drafts the reply; nothing posts until you press Post to Google or Post to FacebookCRM Reputation
AffiliatesNew signups from your affiliate form wait as pending until you press ApproveCRM Affiliates

How do I build an approval gate into an agent?#

Put a Branch in front of the step that acts. Set Field to your approval key, tick Require truthy, and wire the step to the true output. The Review Responder template works exactly this way: approved reviews get a reply, and the rest return on an approval-required output.

Add an approval gate

0 of 4 done

  1. Step 1: Open the Review Responder template

    See a working gate: Input, Branch on approved, the reply step, Slack Message, and a separate approval-required output.
    Open the template
  2. Step 2: Copy the pattern into your agent

    Add a Branch before any step that sends, posts or spends. Point Field at the key that carries the approval.
  3. Step 3: Decide what the false path does

    Return it on its own output, add a task for a reviewer, or post the draft to Slack for a thumbs-up.
  4. Step 4: Test both paths free

    Branches run for real in dry runs. Press Run test once with approval on and once with it off.

How does an AI agent escalate to a human?#

It creates work your team cannot miss. In a CRM workflow, have a Claude step classify the situation, branch on the result, and hand the hard ones to a person with full context.

  • Create Task assigns a call-back with a due time in hours or days, so nothing sits in an inbox.
  • Add Note writes the agent's summary on the contact's timeline, so the person picking it up starts informed.
  • Slack Message pings the channel your team already watches.
  • Send Email to yourself works when the decision belongs to the owner.

What should make an agent escalate?#

Escalate on stakes, not on volume. Write these rules into the classifier step's prompt and route on its answer.

Refund or cancellation requests, a review of 3 stars or lower, a customer who replied twice without an answer, or any legal or safety language.

Questions buyers ask#

What is a human-in-the-loop AI agent?

It is an agent that runs routine work on its own and pauses for a person at the steps you choose. In Agent Studio that means approval gates, draft-then-approve replies and escalation tasks.

Can I approve AI replies before they go out?

Yes. Review replies wait for your Post to Google or Post to Facebook, social posts wait when Needs approval is on, and any agent step can sit behind a Branch that only fires on approval.

How does the agent hand a conversation to my team?

Add Create Task and Add Note to the workflow. The task assigns the follow-up with a due time and the note carries the agent's summary on the contact's timeline.

Does GoHighLevel have approval steps, and what are they in Agent Studio?

GoHighLevel uses workflow If/Else branches and internal notifications. Agent Studio covers the same ground with Branch, Create Task and Slack alerts, plus approval queues in Social and Reputation.

Can I test escalation paths without bothering my team?

Yes. Dry runs are free, branches run for real, and outside steps return simulated results, so you see every path without a task or message reaching anyone.

Was this page helpful?

Good work starts with a clear flow.

Give an agent a job. Connect the steps. See how the work becomes a service someone else can call.

Input
Reason
Output

A workflow you can inspect

Make the steps visible.

Start with a template or describe the job. Connect inputs, reasoning, and output on the canvas.

Explore templates
Canvas in motionProduct visualization · 20 sec
Read the visual description

A stylized organization chart opens into a workflow canvas. Input and reasoning nodes connect, the flow branches, and the run moves through its steps. The film ends with Suede AI Agent Studio. Music only; no spoken narration.